KinetiX AI: fitness app with an AI coach

KinetiX AI is the idea of a fitness coach from Sofia: an app that does what he does in the gym when he stands next to a client. It watches you perform an exercise, counts your reps and tells you what to fix while you’re doing it. Not after the workout, and not from a video you sent somewhere. Right on the phone.

We started in August 2026 with the question that decides everything: can a phone tell the difference between a proper and an improper squat in real time, without a server? We built a browser prototype in two weeks: the camera, a pose model (33 points on the body, 30 times per second) and the first rules for squats and bicep curls. The client opened it on his phone at the gym, recorded the screen and sent us the clips. From there, we worked on his recordings every week: which camera angle is misleading, when a back is “rounded” and when the person is simply short, how late is too late for a cue.

The product’s core is the ai-core package: points from the model go in, and rep-level verdicts and instructions come out. Each exercise has rules for joint angles, tilt and symmetry, with thresholds tuned against real videos, not a textbook. Before analysis starts, the app checks the camera angle and whether the whole body is visible, because half of incorrect verdicts come from a poorly positioned phone. The coach gives one instruction at a time—the most important one—and waits for you to do it; it praises you when your reps are clean. The voice is recorded, not synthesized on the spot: Bulgarian and English, calm, with no exclamation marks.

Around the core, we built the complete app: sign-up with an email code, Google or Apple, onboarding, ready-made workouts by category (HIIT, strength, stretching, technique), a player with sets, rest periods and trainer video, progress with history and clean reps by exercise, and a Pro subscription for AI mode. Everything is bilingual from the first screen.

The backend is Next.js with PostgreSQL: authentication, catalog, sessions, subscription permissions and Stripe. The admin panel is for the client himself: he manages exercises and AI criteria, workout programs, uploads his videos (the server transcodes them with ffmpeg), views users, statistics, payments and the early-access email list from the website. Developers have a separate role because a new exercise requires algorithm work, not just another row in a table.

The kinetixfit.com website follows the product’s brand guidelines: green pose nodes as a visual signature, real recordings from the app instead of stock footage, two languages and an early-access form that writes directly to the backend. We registered the client’s company with the Apple Developer Program and Google Play Console, released the app in TestFlight and set up its business email on its own domain.

The project is in progress: the app is with the client for testing, with the App Store, Google Play and in-store subscriptions still to come.